Direct answer: When an integration fails, capture one real request and response, validate the JSON shape, compare required fields and types, then isolate the smallest payload that reproduces the error. This finds contract drift much faster than changing client code by guesswork.
Most API incidents are described as an application bug even when the application is behaving exactly as it was written. A field changed from a string to an object. A new required parameter appeared. A successful HTTP response wrapped an error in a nested status value. None of those problems are visible if you only look at the final exception in the UI.
Start with a saved request. Record the endpoint, method, headers that affect behavior, query parameters, and body after secrets have been redacted. Do the same for the response: status code, relevant headers, and the complete JSON body. Then run the JSON through a formatter and validator. Indentation is not cosmetic here; it lets you see whether an array became an object, a nullable field disappeared, or a number arrived as a quoted string.
Next, reduce the request. Remove optional fields one at a time until you have the smallest input that still fails. A minimal reproduction tells you whether the problem belongs to authentication, serialization, validation, or the downstream API itself. It also gives a teammate or vendor something they can reproduce without reading your whole codebase.
Finally, turn the observation into a guardrail. Add schema validation at the boundary, log the request ID supplied by the API, and write a regression test using a sanitized fixture. That way the next contract change becomes a clear validation error close to the integration instead of a vague failure several layers later.
GetCodingTools is useful for this loop: format and validate JSON, compare payloads, inspect URL encoding, and generate a clean minimal request without installing another utility.